[MaixCAM2] read_raw() returns an empty buffer (and segfaults on the second call)

[chant555]

Just a heads-up rather than an urgent report — I was trying out raw Bayer capture
on MaixCAM2 and it doesn't seem to work as the API suggests. Not blocking anything
on my side, but you probably want to know.

Environment

MaixCAM2 (2K variant), MaixPy 4.12.5, Ubuntu 22.04 / Linux 4.19.125 aarch64.
This is the latest image available for MaixCAM2 as far as I can tell — v4.12.5
looks like it was a MaixCAM-only release.

What happens

from maix import camera

c = camera.Camera(raw=True)
c.skip_frames(30)
raw = c.read_raw()

Metadata comes back correct — 1280x720, FMT_RGGB10, 1,152,000 bytes
(1.25 B/px, stride 1600, consistent with MIPI RAW10).

But the buffer is empty: 100% zeros, only 4 unique values across 1.15 MB,
mean/std = 0.00 / 0.15.

An RGB frame taken at the same moment is perfectly fine (std 74.3, 256 unique
values), so the sensor itself is clearly producing data.

Separately, asking for a second frame in the same process segfaults:

frame 0: 1152000 B
Segmentation fault (core dumped)

Calling read() before read_raw() triggers it too.

Things I tried

• gc.collect() / del between calls — still segfaults
• format=FMT_RGGB10 in the constructor — "Format not support"
• explicit resolution — "RGB888 must not exceed 1920x1080"
• stopping launcher first — no change
• restarting the process for each frame — no crash, 8 frames captured, but every
  buffer still empty

That last one suggests the empty buffer and the crash are two separate things.

Question

The docs mention RAW output mainly in the astrophotography / long-exposure context.
Is the raw path only wired up for that mode on MaixCAM2? If it needs a particular
setup, that'd be worth a note in the docs — camera.Camera(raw=True) + read_raw()
reads like it should just work.

Happy to test anything if it's useful.

[lxowalle]

Thanks for the feedback, we'll fix it ASAP.